About to upgrade my pc for Terragen 4. Tips ?

Started by Profane, June 05, 2016, 03:34:14 AM

Previous topic - Next topic

AP

#15
Perhaps someday I can upgrade. For the time being all I have is my Ultrabook.

3DnTechNut

I just finished building a new workstation computer for 3D modeling and rendering, Terragen, photo and video processing and it is working well but I went a little crazy in some areas.  My new system build is:

ASSRock x99 Extreme 4 Motherboard
Intel Core i7-5820K 3.3GHZ 6 Core processor (OC'd to 4.3GHZ)
128GB of RipJaws 2400 DDR4 Memory (Probably a bit of overkill but I couldn't resist)
512GB Samsung NVME 950 Pro M.2 SSD (OS and Application Drive)
500GB (x2) Samsung 850 EVO SSD drives
NVIDIA Quadro M4000 Workstation Card (8GB DDR5)
LG Internal Blu-Ray Burner Optical drive
EVGA Supernova 750 Gold PSU
Corsair H100i v2 Hydro Series CPU cooler
6 Cougar 120mm 1200 RPM Silent Case Fans (Intakes)
1 Cougar 120mm 1200 RPM Silent Case Fan (Exhaust)
NZXT Sentry Mix 2 6 Fan Controller
Corsair Vengeance C70 Case to hold everything.

I went this route because the motherboard will support both i7 and Xeon processors and I eventually want to upgrade to a higher grade Xeon processor, though I spent to much on everything else to get the Xeon this year.  :-\

So far I am happy with the results but still trying to tinker with quality, cache and bucket size to decrease my render times and maintain maximum CPU usage through the duration of the render.  I have the system performing pretty well with all 6 cores (12 threads) running at 100% my system temperature is sitting stable at 68 (c) to 72 (c) for the duration of the renders.

I can verify Oshyan's statement about NVIDIA Quadro cards or workstation cards in general not really being an advantage for use with Terragen, though it does help with Adobe Photoshop, Premiere Pro and some other 3D modeling applications.  GeForce cards should suffice nicely.  Oshyan is also spot on about a ton of ram not speeding up renders, but I like having the headroom to handle more complicated scenes if I so desire to over complicate things, which I often do, the net result of watching This is Spinal Tap in my younger formative days. 

Upgrading the CPU and RAM is definitely a good way to go.

Don't be surprised if your upgrades do not cut your render times in half though.  With my graphics laptop running an i7 4700M 3.4GHz processor and 24GB DDR 3 RAM I posted 14:52 for the Terragen Benchmark, with my new system I believe I only saw a 17% to 20% performance increase on the benchmark.  I will run it again and verify.  Don't get me wrong that 17% to 20% increase is still a lot if you have large and long scenes to render, it just most likely won't take your 33 hour render down to 16 hours.

I originally wanted to get a dual Xeon system from a well know workstation company but the cost proved to be prohibitive this year, it was going to run almost $8,000 US, I built this system for less than half.

Now just biding my time to get my hands on Terragen 4 to see how the system handles it. I have the pre-order in and waiting as the days so slowly creep by until it is released.  :)

Kadri

Quote from: 3DnTechNut on June 08, 2016, 04:01:04 AM
...
Don't be surprised if your upgrades do not cut your render times in half though.  With my graphics laptop running an i7 4700M 3.4GHz processor and 24GB DDR 3 RAM I posted 14:52 for the Terragen Benchmark, with my new system I believe I only saw a 17% to 20% performance increase on the benchmark.  I will run it again and verify.  Don't get me wrong that 17% to 20% increase is still a lot if you have large and long scenes to render, it just most likely won't take your 33 hour render down to 16 hours.
...

Nice computer :)

There might other reasons i don't know. Curious.
But i would try to see if 2 or more open seasons of Terragen could bring render times to twice as fast.
It would help for animation  but even with one image you could use crop rendering for example.
It helped me with 32 RAM. Especially with that much RAM you have i would try that.

Profane

Thanks to all of you for your helpful replies. I made my call and i will keep rocking my i5, going to OC it a bit and upgrading my ram to 16gb.
Gaming is still too essential part of my day to day life so getting the new gpu seems to be the best upgrade for me since the tg4 won't be notably heavier to run.

Now begins the long wait for the TG4 release :) !

Oshyan

That sounds like a really sensible upgrade choice indeed. 16GB of RAM will be good for TG4. :)

- Oshyan

3DnTechNut

Quote from: Kadri on June 08, 2016, 06:04:31 AM
Quote from: 3DnTechNut on June 08, 2016, 04:01:04 AM
...
Don't be surprised if your upgrades do not cut your render times in half though.  With my graphics laptop running an i7 4700M 3.4GHz processor and 24GB DDR 3 RAM I posted 14:52 for the Terragen Benchmark, with my new system I believe I only saw a 17% to 20% performance increase on the benchmark.  I will run it again and verify.  Don't get me wrong that 17% to 20% increase is still a lot if you have large and long scenes to render, it just most likely won't take your 33 hour render down to 16 hours.
...

Nice computer :)

There might other reasons i don't know. Curious.
But i would try to see if 2 or more open seasons of Terragen could bring render times to twice as fast.
It would help for animation  but even with one image you could use crop rendering for example.
It helped me with 32 RAM. Especially with that much RAM you have i would try that.

My render times above were from the first day I built the system. I forgot that the day I did the benchmark I had not overclocked the CPU yet and I only had 32GB of Corsair Ballistix Ram in the PC because the motherboard would not post using 8 RAM modules from two 4 module kits, that is why I decided to upgrade to an 8x16Gb module kit.

I did the benchmark again today with the CPU overclocked to 4.4GHz and the DRAM reference clock set to 133mhz with all 12 threads and 100% CPU utilization and the time was 10:50 which is actually a 28% increase from the earlier render time.

I ran two instances of Terragen as you suggested; I did a left and right half crop render of the scene, and put a max of 6 threads on each render.  The two renders finished in 06:00. This was a 45% decrease in render time and yet in both instances all 6 cores and twelve threads were active at 100% CPU utilization.  Confusing.

Even more confusing is that rendering the benchmark with 6 physical cores and no logical cores rendered the entire scene in 08:09 roughly another 25% increase from the benchmark render. The single 6 core render running at 54% CPU utilization took 2 minutes longer to render the entire scene than running 2 instances of the program with each rendering half the scene and 100% CPU utilization.

I tried changing the cache size and rendering with the 6 physical cores versus the 6 cores and 6 hyperthreads and everytime using just the physical cores rendering was about 23% faster, so how can using both the physical and logical cores together in 2 instances be 25% faster then 1 rendering.

I read in the forums before about Terragen not handling hyperthreads as well as physical cores but it seems like it would be true anytime hyperthreads were used.

ajcgi

I thought this virtual cores vs physical cores thing had been sussed? I've just been tearing along with hyperthreading on with nary a care in the world.

3DnTechNut

Just for fun I opened 3 instances of Terragen and did a three way crop, Left Center, Right and it finished in 05:35 about another 10% - 11% increase, interesting! However running three instances did raise system temp by 2 to 3C.  It definitely renders faster on multicore systems if I do split crop renders, but, I don't think I would want to stitch everything together constantly so probably just something to keep in the vault for really big render jobs of still frames.

So this was true in T2 and T3, curious if it is still as significant in T4?  Eagerly waiting to find out.

Oshyan

#23
If you look at the benchmark results list you can see that your results are actually atypical. All of the other 5820k CPUs are *notably* faster than yours. This surprised us too, so we emailed you about it.
https://docs.google.com/spreadsheets/d/1eX9Ltn3_9BjsamA0Pxeflv5AKrjkgViEY8VuetB8e3k/edit?usp=sharing

TG3 (and 4) are not ideal at multithreading, but we have made steady progress on the issue since TG2 and things are very good now for average machines (<32 threads). And both can make reasonable use of hyperthreads (logical rather than physical cores), although at very high core/thread counts (>32), it is best to only use physical cores as the overhead for use of hyperthreads becomes more than the modest performance gain that hyperthreading usually brings (~20%).

- Oshyan

3DnTechNut

#24
Quote from: Oshyan on June 09, 2016, 06:20:41 PM
If you look at the benchmark results list you can see that your results are actually atypical. All of the other 5820k CPUs are *notably* faster than yours. This surprised us too, so we emailed you about it.
https://docs.google.com/spreadsheets/d/1eX9Ltn3_9BjsamA0Pxeflv5AKrjkgViEY8VuetB8e3k/edit?usp=sharing

TG3 (and 4) are not ideal at multithreading, but we have made steady progress on the issue since TG2 and things are very good now for average machines (<32 threads). And both can make reasonable use of hyperthreads (logical rather than physical cores), although at very high core/thread counts (>32), it is best to only use physical cores as the overhead for use of hyperthreads becomes more than the modest performance gain that hyperthreading usually brings (~20%).

- Oshyan

I was wondering about that myself, but if I just load the benchmark file and render without lowering the max cores to 6 that is the result I get every time.

Oshyan

Yes, it's very strange. Something is definitely not right. We'll look into it with you through the support mailbox and hopefully get to the bottom of it.

- Oshyan